Yes, a hypothetical user's sprinker layout "map" or whatever they're working on is actually composed of a few rectangles that represent their house, and a spline representing the garden border, and a circle representing the tree in the front yard, and a bunch of line segments that draw the pipes between the sprinkler heads. Yes, each of those geometric elements can be concisely defined by JSON text that defines the X and Y location, the length/width/diameter/spline coordinates or whatever, the color, etc. of the objects on the map. And yes, OP has a rendering engine that can turn that JSON back into an image.
But when the user thinks about the map, they want to think about the image. If a landscaping customer is viewing a dashboard of all their open projects, OP doesn't want to have to run the rendering engine a dozen times to re-draw the projects each time the page loads just to show a bunch of icons on the screen. They just want to load a bunch of PNGs. You could store two objects on disk/in the database, one being the icon and another being the JSON, but why store two things when you could store one?